CN113194493A - Wireless network data missing attribute recovery method and device based on graph neural network - Google Patents
Wireless network data missing attribute recovery method and device based on graph neural network Download PDFInfo
- Publication number
- CN113194493A CN113194493A CN202110490184.4A CN202110490184A CN113194493A CN 113194493 A CN113194493 A CN 113194493A CN 202110490184 A CN202110490184 A CN 202110490184A CN 113194493 A CN113194493 A CN 113194493A
- Authority
- CN
- China
- Prior art keywords
- attribute
- graph
- matrix
- neural network
- wireless network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 69
- 238000011084 recovery Methods 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 37
- 239000011159 matrix material Substances 0.000 claims abstract description 112
- 239000013598 vector Substances 0.000 claims abstract description 71
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 37
- 238000005070 sampling Methods 0.000 claims abstract description 23
- 238000013507 mapping Methods 0.000 claims abstract description 9
- 238000003062 neural network model Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 10
- 238000004590 computer program Methods 0.000 description 7
- 238000005259 measurement Methods 0.000 description 7
- 210000002569 neuron Anatomy 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/04—Arrangements for maintaining operational condition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a wireless network data loss attribute recovery method and device based on a graph neural network. The method comprises the following steps: mapping wireless network data into a corresponding topological graph structure, and mapping sample data with missing attributes into attribute vectors of nodes in the topological graph structure in sequence; acquiring an adjacency matrix of a topological graph structure according to the attribute vector of the node; simplifying the topological graph structure by using a graph sampling algorithm to obtain a sparse adjacent matrix; and learning by using a graph neural network model based on the attribute vector and the sparse adjacent matrix, and outputting the attribute vector recovered after reconstruction. The method directly models and learns the attribute recovery problem of the wireless network data by using an attribute recovery framework based on a graph automatic encoder and adopting a graph neural network learning algorithm based on a strategy gradient, and makes full use of the correlation in the wireless network data, thereby improving the performance of the attribute of the wireless network data recovery.
Description
Technical Field
The invention relates to the problem of attribute recovery of wireless network data, in particular to a method and a device for recovering missing attributes of wireless network data based on a graph neural network.
Background
Machine learning and deep learning have achieved tremendous success in the past few years, and while new technological breakthroughs continue to emerge, most methods of supervised learning still require data sets with complete information. At the same time, many real-world problems still require processing data sets with incomplete information, such as biomedical or, insurance departments, or financial institutions. Therefore, the need to supplement the incomplete data set with the complete operation is an essential part of machine learning.
The attribute recovery algorithm is used for acting on data sets with missing data, and a specific algorithm is used for replacing and supplementing missing parts with predicted values of the algorithm to finally obtain a complete data set.
The wireless network data attribute recovery is attribute recovery performed on a data set having a wireless network structure, and the attributes possessed by the data of the wireless network generally include the geographical location of a wireless base station, the geographical location of a mobile terminal, device information of the mobile terminal, communication signal strength of the mobile terminal, and the like. In a large wireless network environment consisting of a large number of wireless network infrastructure and mobile terminals, there are many tasks that need to be performed depending on the data attributes in the wireless network environment, such as wireless network measurements, which are important for operators and researchers to know the network performance, to evaluate the quality of experience of users, and to facilitate the deployment of infrastructure and resources. The measurement of the wireless network depends on the data attributes of the wireless network infrastructure and the mobile terminal in the scene, and the wireless network measurement can be effectively carried out only if the complete data attributes are obtained. However, in an actual network environment, it is basically impossible to obtain complete data attributes, and due to the characteristics of wireless communication, a series of difficulties exist to cause the data attributes to be lost, so that the wireless network measurement task cannot be performed. Therefore, attribute recovery of wireless network data is required to supplement a data set of a complete network, so that a subsequent wireless network measurement task can be effectively supported.
There are some places where wireless network data is different from other sample data sets, and there may be some correlation between sample data in these data sets, and this correlation brings certain challenges to attribute recovery of wireless network data, because it is necessary to make good use of the correlation between such sample data, which may help the accuracy of attribute recovery, but how to apply the correlation between such sample data to attribute recovery is a problem that needs to be considered and solved.
At present, in an attribute recovery algorithm, an attribute recovery problem is constructed as a prediction task, and variants of a standard supervision algorithm can be applied on the basis of the prediction task, wherein the variants comprise K-NN, decision trees, support vector technology and the like. But K-NN is limited in performing a weighted average of similar feature vectors, while other algorithms require the creation of a global model of the data set for the computation.
In recent years, there has been a proliferation of interest in applying deep learning techniques to the attribute recovery problem. The method comprises multiple subsumption of a depth denoising autoencoder, combination of a depth network and a probability mixed model, and a variational autoencoder. In general, these methods better capture complex correlations in data because they have multiple levels of non-linear computations, but they still require the construction of a global model from the data set while ignoring potentially significant contributions from similar attributes.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the problems, the invention provides a wireless network data missing attribute recovery method based on a graph neural network, which can fundamentally solve the problem that the existing wireless network data attribute recovery algorithm cannot effectively utilize the dependency of related attribute information to cause the performance of recovery attribute to be not high enough.
The invention also aims to provide a wireless network data missing attribute recovery device based on the graph neural network.
The technical scheme is as follows: in order to achieve the above object, the technical solution of the present invention is as follows:
in a first aspect, a method for recovering data missing attributes of a wireless network based on a graph neural network is provided, which includes the following steps:
(1) mapping the wireless network data with the missing attributes into attribute vectors of graph nodes of the corresponding topological graph structure to obtain an attribute matrix of the topological graph structure;
(2) acquiring an adjacency matrix of the topological graph structure by utilizing a graph structure learning algorithm based on the attribute matrix of the topological graph structure;
(3) simplifying the topological graph structure by using a graph sampling algorithm to obtain a sparse adjacent matrix;
(4) and inputting the attribute matrix of the topological graph structure and the sparse adjacent matrix into a neural network of an automatic graph encoder to obtain a reconstructed attribute vector, wherein the neural network of the automatic graph encoder comprises an encoder and a decoder, the encoder encodes the input vector by using a convolutional graph neural network to obtain an intermediate vector, and the decoder decodes the intermediate vector by using a full connection layer to output the reconstructed attribute vector.
As a preferred embodiment, the method further comprises the steps of:
(5) and updating parameters in the graph structure learning algorithm and the graph automatic encoder neural network according to the error between the reconstructed vector and the real complete attribute vector output by the graph automatic encoder neural network.
In a second aspect, a wireless network data loss attribute recovery device based on a graph neural network is provided, which includes:
the graph structure establishing module is used for mapping the wireless network data with the missing attributes into the attribute vectors of the graph nodes of the corresponding topological graph structure to obtain an attribute matrix of the topological graph structure;
the graph structure learning module is used for acquiring an adjacent matrix of the topological graph structure by utilizing a graph structure learning algorithm based on the attribute matrix of the topological graph structure;
the graph structure sampling module is used for simplifying the topological graph structure by utilizing a graph sampling algorithm to obtain a sparse adjacent matrix;
the automatic graph encoder neural network module is used for obtaining a reconstructed attribute vector by using the automatic graph encoder neural network according to the attribute matrix of the topological graph structure and the sparse adjacent matrix, the automatic graph encoder neural network module comprises an encoder unit and a decoder unit, the encoder unit encodes an input vector by using the convolutional graph neural network to obtain an intermediate vector, the decoder unit decodes the intermediate vector by using the full connection layer, and the output vector is the reconstructed attribute vector.
As a preferred embodiment, the wireless network data loss attribute recovery device based on the graph neural network further includes: and the learning updating module is used for updating parameters in the graph structure learning algorithm and the graph automatic encoder neural network according to the error between the reconstructed vector output by the graph automatic encoder neural network and the real complete attribute vector.
Has the advantages that: the invention firstly proposes to use the graph neural network model to solve the attribute recovery problem of the wireless network data, and uses the graph automatic encoder to reconstruct the missing attribute vector, thereby fundamentally solving the problem that the prior wireless network data attribute recovery algorithm cannot effectively utilize the dependency of the related attribute information to cause the property recovery performance to be not high enough. In addition, a graph structure self-learning algorithm, a graph structure sampling algorithm and a graph neural network learning algorithm based on strategy gradients are used for directly modeling and learning the attribute recovery problem of the wireless network data, so that the performance of the attribute of the wireless network data recovery is improved.
Drawings
FIG. 1 is a flow chart of a wireless network data loss attribute recovery method based on a graph neural network;
FIG. 2 is an exemplary diagram of a neural network of the graph autoencoder according to an embodiment of the present invention;
fig. 3 is a block diagram of a wireless network data missing attribute recovery device based on a graph neural network.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings. It should be understood that the following embodiments are provided only for the purpose of thoroughly and completely disclosing the present invention and fully conveying the technical concept of the present invention to those skilled in the art, and the present invention may be embodied in many different forms and is not limited to the embodiments described herein. The terminology used in the exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting of the invention.
FIG. 1 is a flow chart of a wireless network data missing attribute recovery method based on a graph neural network, and as shown in the figure, the invention provides a new graph automatic encoder neural network model with a self-learning graph structure to solve the attribute recovery problem of wireless network data. A wireless network data missing attribute recovery method based on a graph neural network comprises the following steps:
(1) and mapping the wireless network data with the missing attributes into the attribute vectors of the graph nodes of the corresponding topological graph structure to obtain an attribute matrix of the topological graph structure.
In a wireless network, data describing a mobile terminal and a base station are acquired by manual collection, wherein the data includes location information data of the device and the base station, such as absolute location information of the mobile terminal and the base station, including longitude and latitude, relative location information including whether the mobile terminal and the base station are located indoors or outdoors, which room the mobile terminal and the base station are located in, and the like, signal strength measurement values between the mobile terminal and the base station, and use time data of the base station, and the like. The data used for describing the situation of each entity in the wireless network is defined as the attribute of each entity, and if some data in the data has blank and is not recorded, for example, the signal strength value between a certain mobile terminal and a base station shows blank, the attribute of the mobile terminal is considered to have missing. In the data describing the base station and the mobile terminal of the wireless network, the position information data of the mobile terminal and the base station and the signal strength data between the mobile terminal and the base station have an obvious relationship, and if the position information of the base station and the mobile terminal is similar, the signal strength data of the mobile terminal and the base station are larger. And finally, mapping the mobile terminal and the base station of the wireless network into nodes in a graph structure, wherein the attributes attached to the mobile terminal and the base station are the attributes of the nodes of the corresponding graph.
For N wireless network data samples, under the condition that each sample has D attribute dimensions, the dimension of the finally obtained attribute matrix is NxD. Here, all data have D attributes, for example, in a wireless network, each mobile terminal has four attribute dimensions of a device number, a device longitude value, a device dimension value, and a device signal strength value, and then an attribute matrix dimension regarding the mobile terminal data is N × 4. It should be understood that the data illustrated herein is for illustrative purposes only and is not intended to limit the present invention to four attribute dimensions, and that there may be other different corresponding attributes for data acquired for different measurement purposes.
(2) And acquiring an adjacency matrix of the topological graph structure by utilizing a graph structure learning algorithm based on the attribute matrix of the topological graph structure.
The graph structure learning algorithm is realized according to a set self-learning topological graph structure formula, the attribute matrix is converted into a potential space by using a conversion matrix to be learned, and then the spatial distance of the attribute vector in the potential space is calculated to obtain an adjacent matrix which can reflect the correlation of the attribute vector of the graph node and is related to the topological graph structure.
The conversion matrix is two parameter matrices, the optimal parameter value needs to be determined through learning, random initialization can be performed at the beginning of learning, and the determined optimal parameter value can be directly used in application after the learning stage is completed. The two transformation matrices each map the attribute matrix to a different matrix in the underlying space. Let the attribute matrix be X, and set two transformation matrices to be learned as theta respectively1And Θ2The conversion formula for converting the attribute matrix into the latent space is Si＝tanh(XΘi),i＝1,2。
The formula for calculating the adjacency matrix of the topological graph structure is as follows:
wherein, theta1And Θ2Is the wholeConverting parameters which need to be obtained through training are unified in the end-to-end model, and the original attribute matrix X is respectively converted into a new matrix S through the two parameters1And S2By matrix multiplication of the two new matrices, an inner product between two vectors, that is, a spatial distance between the two vectors, can be obtained, and the purpose of subtraction is to obtain a heterogeneous adjacency matrix.
(3) And simplifying the topological graph structure by using a graph sampling algorithm to obtain a sparse adjacent matrix.
In the graph sampling algorithm, the sampling times are set, the edge which should be reserved by each node is calculated, the reservation mode is that row vectors in an adjacent matrix corresponding to the nodes are sorted from large to small to obtain corresponding index values, only the maximum elements with the same number as the sampling times are reserved, and the rest elements are set to be zero.
The reason for performing the graph sampling is: the graph adjacency matrix obtained through the self-learning graph structure is too dense, and if a large number of elements close to 0 are contained, if the elements are reserved, the topological graph structure is too complex, each node has a large number of neighborhood nodes, so that the neighborhood nodes of different nodes have a large number of same nodes, and when the neighborhood nodes of the nodes cannot embody local characteristics, the effect of the graph neural network model is greatly reduced. Because when too many neighborhood nodes participate in convolution, the nodes corresponding to small weight values reduce the influence of important neighbor nodes on the central node, the convolution result tends to make the embedded vectors of most nodes consistent, and moreover, a dense adjacency matrix also consumes a large amount of calculation cost. The density of the adjacency matrix needs to be reduced, which not only helps to reduce the time cost of graph convolution operation, but also can reduce the interference of the joint-free point to the central node and enhance the convolution effect of the key neighbor nodes on the central node. In the sampling algorithm of the graph structure, for each node, K nodes with the maximum weight in the adjacent matrix are reserved as neighborhood nodes of the central node. Only K elements are retained per row vector in the adjacency matrix, with the other elements set to zero. K ranges from 2 to 6.
The graph structure sampling algorithm of the embodiment of the invention is as follows:
the algorithm mainly comprises the following steps: for each row vector in the adjacency matrix, counting the maximum K elements in the row vector, keeping the numerical values of the K elements unchanged, setting the numerical values of the rest N-K elements to be 0, and performing the operation on each row by the analogy to finally obtain a new adjacency matrix which is a sparse adjacency matrix.
(4) And inputting the attribute matrix of the topological graph structure and the sparse adjacent matrix into a neural network of an automatic encoder of the graph to obtain the reconstructed attribute vector.
After the sparse adjacency matrix is obtained, the existing attribute matrix and adjacency matrix of the graph nodes are required to be used for constructing a graph automatic encoder neural network module. The graph autoencoder neural network according to the embodiment of the present invention is shown in fig. 3, and for convenience of description, the graph autoencoder neural network is also referred to as a model hereinafter. The model is divided into an encoder and a decoder, the encoder encodes input vectors (including attribute matrixes of nodes and sparse adjacent matrixes) to obtain hidden vectors belonging to intermediate states of the nodes, and in order to extract spatial information of a graph structure, the model can use a graph convolution neural network to construct the encoder, so that the nodes can fully integrate information of neighborhood nodes in the encoding process. Wherein the encoder is defined as:
where σ and W are the activation function and the parameters that need to be trained in the end-to-end model,
i denotes a unit matrix of the cell,
is based on
The corresponding degree matrix is obtained. According to the formula, the encoder multiplies the degree matrix and the adjacent matrix by the attribute matrix after the function, and then multiplies a parameter matrix to obtain a final matrix H, wherein the matrix H is the encoding result of the encoder.
After obtaining the intermediate state vector of the node, it is necessary to reconstruct the intermediate vector containing the missing attribute into the attribute vector of the node using a decoder constructed using a fully-connected layer. In the decoder part, the obtained matrix H is input into a full-connection neural network, the matrix H is changed into a reconstructed attribute matrix which has the same dimension with the original attribute matrix through the parameter action of neurons in two hidden layers and the action of an activation function, and elements of missing positions in the matrix H are filled in the complete reconstructed attribute matrix.
The specific parameters of the neural network in this embodiment are: in the encoder, the convolutional neural network has a hidden layer, the number of neurons ranges from 16 to 64, each neuron uses a ReLU activation function, in the decoder, the fully-connected neural network has two hidden layers, the number of neurons also ranges from 16 to 64, and each neuron uses a ReLU activation function. The learning rate of the neural network is 0.001.
On the basis of the prior art, the invention firstly proposes to use the learning parameters to construct a sparse adjacency matrix which can provide better support for the recovery of the missing attribute, solves the problem of the recovery of the missing attribute with a new angle, redefines the recovery process of the attribute by using a framework of automatic coding of a graph, improves a coding formula in the coding process, introduces the learned adjacency matrix, and can fundamentally solve the problem that the performance of the recovery attribute is not high enough because the prior wireless network data attribute recovery algorithm cannot effectively utilize the dependency of related attribute information.
(5) And updating parameters in the graph structure learning algorithm and the graph automatic encoder neural network according to the error between the reconstructed vector and the real complete attribute vector output by the graph automatic encoder neural network.
And performing loss function error calculation by using the reconstructed node vector and the real complete attribute vector, calculating an error gradient, reversely propagating the gradient, updating network parameters of a decoder and an encoder in a neural network of the automatic encoder of the graph, and further updating learning parameters of a self-learning network structure. The above process is repeated until the error reaches an optimum value.
According to another embodiment of the present invention, as shown in fig. 3, a wireless network data loss attribute recovery apparatus based on a graph neural network is provided, which mainly includes a graph construction module, a graph structure learning module, a graph structure sampling module, a graph automatic encoder neural network module, and a learning update module.
The graph building module is used for mapping the wireless network data with the missing attributes into the attribute vectors of the graph nodes of the corresponding topological graph structure to obtain the attribute matrix of the topological graph structure. For N wireless network data samples, under the condition that each sample has D attribute dimensions, the dimension of the finally obtained attribute matrix is NxD.
The graph structure learning module can learn the topological graph structure of the wireless network data by self, and comprises the step of extracting an adjacency matrix from the wireless network data, wherein elements in the adjacency matrix have learned weight values, so that the importance of neighborhood nodes to a central node can be effectively reflected, and the subsequent graph automatic encoder neural network module is allowed to integrate the information of the neighborhood nodes according to the importance of the neighborhood nodes. Specifically, the graph structure learning module converts the attribute matrix into a potential space by using the conversion matrix according to a set self-learning topological graph structure formula, and then calculates the spatial distance of the attribute vector in the potential space to obtain an adjacent matrix which can reflect the correlation of the attribute vector of the graph node and is related to the topological graph structure. The calculation method for converting the attribute matrix into the potential space comprises the following steps: si＝tanh(XΘi),i＝1,2，ΘiFor the transformation matrix, there are two transformation matrices in total, and X is an attribute matrix. The formula for calculating the adjacency matrix of the topological graph structure is as follows:
S1and S2Respectively, using the transformation matrix theta1And Θ2And (5) obtaining a new matrix after the attribute matrix X is changed.
The graph sampling module samples the neighborhood nodes of each node based on the learned adjacent matrix with the weight according to the weight to obtain a relatively sparse graph structure, and aims to reduce the interference of relatively unimportant neighborhood nodes on the attribute recovery of the central node. Specifically, the graph sampling module calculates an edge that each node should reserve according to a preset sampling frequency, the reserving mode is that row vectors in an adjacent matrix corresponding to the nodes are sorted from large to small to obtain corresponding index values, only the maximum elements with the same number as the sampling frequency are reserved, and the rest elements are set to be zero.
The graph automatic encoder neural network module uses the graph convolutional neural network to construct an encoder part and a decoder part, and obtains the intermediate vector after node encoding and the reconstructed vector after decoding respectively. Specifically, the graph autoencoder neural network encoder is of the form:
wherein, sigma represents an activation function, A is an adjacency matrix of a topological graph structure, D is an attribute dimension of data,
i denotes a unit matrix of the cell,
is based on
And obtaining a corresponding degree matrix, wherein X is an attribute matrix, W is a neural network model parameter, and the decoder adopts a fully-connected neural network with two hidden layers.
After the reconstructed vector is obtained, the learning updating module calculates the error between the original node attribute and the restored node attribute through a predefined unsupervised reconstruction error loss function, and the gradient is propagated reversely through a gradient descent algorithm to train and optimize the parameters needing to be learned in the whole model framework.
It should be understood that the wireless network data loss attribute recovery apparatus for a neural network in the present embodiment may implement all technical solutions in the foregoing method embodiments, and the functions of each functional module may be implemented specifically according to the method in the foregoing method embodiments, and the specific implementation process may refer to the relevant description in the foregoing embodiments, which is not described herein again.
Based on the same technical concept as the method embodiment, according to another embodiment of the present invention, there is provided a computer apparatus including: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, which when executed by the processors implement the steps in the method embodiments.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (10)
1. A wireless network data missing attribute recovery method based on a graph neural network is characterized by comprising the following steps:
(1) mapping the wireless network data with the missing attributes into attribute vectors of graph nodes of the corresponding topological graph structure to obtain an attribute matrix of the topological graph structure;
(2) acquiring an adjacency matrix of the topological graph structure by utilizing a graph structure learning algorithm based on the attribute matrix of the topological graph structure;
(3) simplifying the topological graph structure by using a graph sampling algorithm to obtain a sparse adjacent matrix;
(4) and inputting the attribute matrix of the topological graph structure and the sparse adjacent matrix into a neural network of an automatic graph encoder to obtain a reconstructed attribute vector, wherein the neural network of the automatic graph encoder comprises an encoder and a decoder, the encoder encodes the input vector by using a convolutional graph neural network to obtain an intermediate vector, and the decoder decodes the intermediate vector by using a full connection layer to output the reconstructed attribute vector.
2. The method for recovering the missing attribute of the wireless network data based on the neural network of the figure as claimed in claim 1, wherein the dimension of the attribute matrix in the step (1) is nxd, where N is the number of data of the wireless network data and D is the attribute dimension of the data.
3. The method for recovering the data missing attribute of the wireless network based on the graph neural network as claimed in claim 1, wherein the graph structure learning algorithm in the step (2) utilizes the transformation matrix to transform the attribute matrix into the latent space according to the set self-learning topological graph structure formula, and then calculates the spatial distance of the attribute vector in the latent space to obtain an adjacent matrix which can reflect the correlation of the attribute vector of the graph node and is related to the topological graph structure.
4. The method for recovering the data missing attribute of the wireless network based on the graph neural network as claimed in claim 3, wherein the calculation manner for converting the attribute matrix into the latent space is as follows: si＝tanh(XΘi),i＝1,2，ΘiFor the transformation matrix, there are two transformation matrices in total, and X is an attribute matrix.
5. The method for recovering the data missing attribute of the wireless network based on the graph neural network as claimed in claim 4, wherein the formula for calculating the adjacency matrix of the topological graph structure is as follows:
S1and S2Respectively, using the transformation matrix theta1And Θ2And (5) obtaining a new matrix after the attribute matrix X is changed.
6. The method for recovering the wireless network data missing attribute based on the graph neural network according to claim 1, wherein the graph sampling algorithm in the step (3) calculates the edge that each node should reserve according to the preset sampling times, the reserving mode is that row vectors in an adjacent matrix corresponding to the nodes are sorted from large to small to obtain corresponding index values, only the maximum elements with the same number as the sampling times are reserved, and the rest elements are set to be zero.
7. The method for recovering the data missing attribute of the wireless network based on the graph neural network as claimed in claim 1, wherein in the step (4), the graph automatic encoder is in the form of a neural network encoder as follows:
wherein, sigma represents an activation function, A is an adjacency matrix of a topological graph structure, D is an attribute dimension of data,
i denotes a unit matrix of the cell,
is based on
And obtaining a corresponding degree matrix, wherein X is an attribute matrix, W is a neural network model parameter, and the decoder adopts a fully-connected neural network with two hidden layers.
8. The method for recovering the data missing attribute of the wireless network based on the graph neural network as claimed in claim 1, wherein the method further comprises: (5) and updating parameters in the graph structure learning algorithm and the graph automatic encoder neural network according to the error between the reconstructed vector and the real complete attribute vector output by the graph automatic encoder neural network.
9. A wireless network data missing attribute recovery device based on a graph neural network is characterized by comprising the following components:
the graph structure establishing module is used for mapping the wireless network data with the missing attributes into the attribute vectors of the graph nodes of the corresponding topological graph structure to obtain an attribute matrix of the topological graph structure;
the graph structure learning module is used for acquiring an adjacent matrix of the topological graph structure by utilizing a graph structure learning algorithm based on the attribute matrix of the topological graph structure;
the graph structure sampling module is used for simplifying the topological graph structure by utilizing a graph sampling algorithm to obtain a sparse adjacent matrix;
the automatic graph encoder neural network module is used for obtaining a reconstructed attribute vector by using the automatic graph encoder neural network according to the attribute matrix of the topological graph structure and the sparse adjacent matrix, the automatic graph encoder neural network module comprises an encoder unit and a decoder unit, the encoder unit encodes an input vector by using the convolutional graph neural network to obtain an intermediate vector, the decoder unit decodes the intermediate vector by using the full connection layer, and the output vector is the reconstructed attribute vector.
10. The wireless network data missing attribute recovery device based on graph neural network of claim 9, further comprising: and the learning updating module is used for updating parameters in the graph structure learning algorithm and the graph automatic encoder neural network according to the error between the reconstructed vector output by the graph automatic encoder neural network and the real complete attribute vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110490184.4A CN113194493B (en) | 2021-05-06 | 2021-05-06 | Wireless network data missing attribute recovery method and device based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110490184.4A CN113194493B (en) | 2021-05-06 | 2021-05-06 | Wireless network data missing attribute recovery method and device based on graph neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113194493A true CN113194493A (en) | 2021-07-30 |
| CN113194493B CN113194493B (en) | 2023-01-06 |
Family
ID=76983711
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110490184.4A Active CN113194493B (en) | 2021-05-06 | 2021-05-06 | Wireless network data missing attribute recovery method and device based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113194493B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113988464A (en) * | 2021-11-17 | 2022-01-28 | 国家电网有限公司客户服务中心 | Network link attribute relation prediction method and equipment based on graph neural network |
| CN114611590A (en) * | 2022-03-01 | 2022-06-10 | 浙江大学 | Graph neural network-based power system missing data reconstruction method and system |
| CN114662204A (en) * | 2022-04-07 | 2022-06-24 | 清华大学 | Elastic bar system structure system data processing method and device based on graph neural network |
| CN115801549A (en) * | 2023-01-28 | 2023-03-14 | 中国人民解放军国防科技大学 | Adaptive network recovery method, device and equipment based on key node identification |
| CN117557118A (en) * | 2023-11-13 | 2024-02-13 | 国网江苏省电力有限公司镇江供电分公司 | UPS system power supply topological graph generation method based on machine learning |
| CN118039003A (en) * | 2024-01-10 | 2024-05-14 | 浙江大学 | Silicon content prediction method for digital twin system of blast furnace based on distribution and graph convolution |
| CN118409734A (en) * | 2024-06-27 | 2024-07-30 | 之江实验室 | Sparse matrix operation programming method and device based on data stream |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110717617A (en) * | 2019-09-09 | 2020-01-21 | 广东工业大学 | Unsupervised relation prediction method based on depth map network self-encoder |
| CN110781406A (en) * | 2019-10-14 | 2020-02-11 | 西安交通大学 | Social network user multi-attribute inference method based on variational automatic encoder |
| CN111950594A (en) * | 2020-07-14 | 2020-11-17 | 北京大学 | Unsupervised graph representation learning method and unsupervised graph representation learning device on large-scale attribute graph based on sub-graph sampling |
| CN112567355A (en) * | 2018-09-04 | 2021-03-26 | 北京京东尚科信息技术有限公司 | End-to-end structure-aware convolutional network for knowledge base completion |
-
2021
- 2021-05-06 CN CN202110490184.4A patent/CN113194493B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112567355A (en) * | 2018-09-04 | 2021-03-26 | 北京京东尚科信息技术有限公司 | End-to-end structure-aware convolutional network for knowledge base completion |
| CN110717617A (en) * | 2019-09-09 | 2020-01-21 | 广东工业大学 | Unsupervised relation prediction method based on depth map network self-encoder |
| CN110781406A (en) * | 2019-10-14 | 2020-02-11 | 西安交通大学 | Social network user multi-attribute inference method based on variational automatic encoder |
| CN111950594A (en) * | 2020-07-14 | 2020-11-17 | 北京大学 | Unsupervised graph representation learning method and unsupervised graph representation learning device on large-scale attribute graph based on sub-graph sampling |
Non-Patent Citations (2)
| Title |
|---|
| HAN ZHANG等: "ReLeS: A Neural Adaptive Multipath Scheduler based on Deep Reinforcement Learning", 《IEEE INFOCOM 2019 - IEEE CONFERENCE ON COMPUTER COMMUNICATIONS》 * |
| 郑昕韬: "基于图神经网络的网络属性预测算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113988464A (en) * | 2021-11-17 | 2022-01-28 | 国家电网有限公司客户服务中心 | Network link attribute relation prediction method and equipment based on graph neural network |
| CN114611590A (en) * | 2022-03-01 | 2022-06-10 | 浙江大学 | Graph neural network-based power system missing data reconstruction method and system |
| CN114662204A (en) * | 2022-04-07 | 2022-06-24 | 清华大学 | Elastic bar system structure system data processing method and device based on graph neural network |
| CN115801549A (en) * | 2023-01-28 | 2023-03-14 | 中国人民解放军国防科技大学 | Adaptive network recovery method, device and equipment based on key node identification |
| CN115801549B (en) * | 2023-01-28 | 2023-06-16 | 中国人民解放军国防科技大学 | Self-adaptive network recovery method, device and equipment based on key node identification |
| CN117557118A (en) * | 2023-11-13 | 2024-02-13 | 国网江苏省电力有限公司镇江供电分公司 | UPS system power supply topological graph generation method based on machine learning |
| CN118039003A (en) * | 2024-01-10 | 2024-05-14 | 浙江大学 | Silicon content prediction method for digital twin system of blast furnace based on distribution and graph convolution |
| CN118409734A (en) * | 2024-06-27 | 2024-07-30 | 之江实验室 | Sparse matrix operation programming method and device based on data stream |
| CN118409734B (en) * | 2024-06-27 | 2024-10-11 | 之江实验室 | Sparse matrix operation programming method and device based on data stream |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113194493B (en) | 2023-01-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113194493B (en) | Wireless network data missing attribute recovery method and device based on graph neural network | |
| US20210133536A1 (en) | Load prediction method and apparatus based on neural network | |
| CN109344992B (en) | Modeling method for user control behavior habits of smart home integrating time-space factors | |
| CN112766600B (en) | Urban area crowd flow prediction method and system | |
| CN116030537B (en) | Three-dimensional human body posture estimation method based on multi-branch attention-seeking convolution | |
| CN117454495B (en) | CAD vector model generation method and device based on building sketch outline sequence | |
| CN115545334B (en) | Land utilization type prediction method and device, electronic equipment and storage medium | |
| CN115001937B (en) | Smart city Internet of things-oriented fault prediction method and device | |
| CN115759413A (en) | Meteorological prediction method and device, storage medium and electronic equipment | |
| CN116401949A (en) | Distributed photovoltaic resource output curve deduction method, system, equipment and medium | |
| CN115758643A (en) | Network flow prediction method and device based on temporal-spatial feature fusion and storage medium | |
| CN118036822A (en) | Electric hydrogen load prediction method, device and storage medium of electric hydrogen integrated energy station | |
| CN117914690A (en) | Edge node network fault prediction method based on deep learning GCN-LSTM | |
| CN117634554A (en) | Space-time deep learning runoff prediction method and system based on spatial information fusion | |
| CN117853596A (en) | Unmanned aerial vehicle remote sensing mapping method and system | |
| CN115604131B (en) | Link flow prediction method, system, electronic device and medium | |
| CN116826734A (en) | Photovoltaic power generation power prediction method and device based on multi-input model | |
| CN110020379B (en) | Link prediction method based on deep dynamic network embedded representation model | |
| CN116303961A (en) | Knowledge graph question-answering method and system based on multi-hop combined graph fingerprint network | |
| CN114445692A (en) | Image recognition model construction method and device, computer equipment and storage medium | |
| CN113360772A (en) | Interpretable recommendation model training method and device | |
| CN118279113B (en) | Digital intelligence social public service management method and system based on large model | |
| CN118114057B (en) | Model training method, runoff prediction method, electronic device, and storage medium | |
| CN118393973B (en) | Automatic driving control method, device, system, equipment and storage medium | |
| CN117939518A (en) | Mobile network usage data prediction method and device suitable for base station |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |